DeepSeek V4.1 Flash refreshed (1017 build) climbs to #2 open-weight on Artificial Analysis Intelligence Index

2026-10-17

Per Artificial Analysis data relayed by Latent Space's AI News digest, an updated build of DeepSeek V4.1 Flash (tagged '1017') scored 45 on the Artificial Analysis Intelligence Index, up 6 points, tying GLM-5.3 for #2 among open-weight models (one point behind MiMo-V2.6-Pro). Its GDPval-AA Elo rose to 1746 (+146) and its AA-Omniscience non-hallucination rate rose from 4% to 33%. Together said DeepSeek trained the update using self-feedback reinforcement learning for building complete projects from scratch. Pricing is unchanged at $0.30/$1.20 per million input/output tokens.

Significance 2: An incremental benchmark refresh of an existing open-weight model via independent measurement (Artificial Analysis) — notable context for the open-weights field but not a frontier-moving or verdict-changing jump.

technology Benchmark result Open weights

Sources

curator [AINews] not much happened today retrieved 2026-10-20

← Back to feed